Research On Denoising Algorithm Of Terahertz Image By Curvelet Transform With Automatic Noise Estimation

Terahertz imaging is one of the important research directions of terahertz technology.With its high penetration and low energy characteristics,terahertz light waves are widely used in fields such as biomedicine,in vivo inspection,human security inspection and non-destructive inspection.At present,terahertz imaging technology has become a research hotspot in the imaging field.However,due to factors such as low signal source energy,unstable radiation source,detector resolution performance limitations and experimental device interference,large image noises are often inevitably generated during the imaging process,resulting in poor terahertz image quality and affecting subsequent Research,so it is necessary to denoise the terahertz image.Since the real terahertz image noise level is unknown,this parameter is required in some denoising algorithms commonly used at present to obtain a better image denoising effect.To this end,a composite noise level estimation algorithm is used to estimate the noise level of the terahertz image.The composite noise level estimation algorithm is a noise level estimation method based on weak texture blocks and local statistics.This algorithm is suitable for terahertz images.Using this estimate,the threshold denoising algorithm of curvelet transform is used to study the denoising of multiple real terahertz images.The effects of threshold algorithms such as Monte Carlo hard threshold,Monte Carlo soft threshold,adaptive hard threshold,adaptive soft threshold,adaptive soft and hard compromise threshold and construction threshold function method in the denoising of curvelet threshold are studied respectively.Aiming at the terahertz real image with extremely low signal-to-noise ratio and periodic noise in the background—digital "100" image,the background noise intensity of the image is too high,and the noise level estimation result is not accurate enough,the notch filter is used to achieve preprocessing and then the curve wave Change the threshold denoising method.Aiming at the real terahertz image with random ring-shaped fringe noise in the background-the letter "G" image,the connected domain segmentation algorithm is used to achieve image preprocessing,remove part of the background random noise,and improve the image quality.Finally,the curvelet transform algorithm is combined with the guided filtering algorithm,and on the basis of the curvelet transform denoising,the real terahertz image is denoised by guided filtering.For the guided filter denoising algorithm,the influence of the guided image and the input image on the output result of the filter is discussed separately.The experimental results show that the guided filtering algorithm has good applicability for denoising terahertz images.